import numpy as np
import pandas as pd
from matplotlib import pyplot as plt
from statistics import mean, median
from math import sqrt
from scipy.stats import mannwhitneyu
from typing import Tuple, List, Dict, Set, Iterable
import matplotlib.style as mpl_style
import os
import json

# path to csv results
EC_PATH = "./results.csv"

def df_from_path(path: str) -> pd.DataFrame:
    return pd.read_csv(
        filepath_or_buffer=path,
        sep="?",
    )

EC_RESULTS = df_from_path(EC_PATH)
EC_RESULTS.tail()

	directory	rund_id	test_no	generation	after_selection_depth_25percentile	after_selection_depth_50percentile	after_selection_depth_75percentile	after_selection_depth_algorithm	after_selection_depth_avg	after_selection_depth_avg_lev_distance_denoising	...	training_error	training_errors	training_mode	unique	unique_output_vector_rate	unique_output_vector_rate_int	unique_output_vector_rate_sel	unique_output_vector_rate_test	unique_rate	wass_norm_lev_div_sampled_vs_selected
615	dae_gp	6_8335	6	26	0.0	0.0	0.0	DAE_LSTM	0.188	0.070	...	0.151	None	convergence	31	0.060	0.052	0.058	0.060	0.062	0.127
616	dae_gp	6_8335	6	27	0.0	0.0	0.0	DAE_LSTM	0.022	0.064	...	0.371	None	convergence	46	0.078	0.072	0.048	0.078	0.092	0.04
617	dae_gp	6_8335	6	28	0.0	0.0	0.0	DAE_LSTM	0.114	0.066	...	0.166	None	convergence	79	0.122	0.100	0.058	0.122	0.158	0.085
618	dae_gp	6_8335	6	29	0.0	0.0	0.0	DAE_LSTM	0.174	0.072	...	0.142	None	convergence	84	0.110	0.092	0.076	0.110	0.168	0.163
619	dae_gp	6_8335	6	30	0.0	0.0	0.0	DAE_LSTM	0.188	0.064	...	0.23	None	convergence	51	0.078	0.072	0.072	0.078	0.102	0.007

5 rows × 787 columns

def split_df(df: pd.DataFrame, dir1: str, dir2: str) -> Tuple[pd.DataFrame, pd.DataFrame]:
    return df.query("directory == @dir1").copy(), df.query("directory == @dir2").copy()
    
pt_results, reg_results = split_df(EC_RESULTS, "pt_dae_gp", "dae_gp")

print(
    pt_results.shape == reg_results.shape,
    pt_results.shape,
    reg_results.shape
)

True (310, 787) (310, 787)

def filter_df_by_headers(df, headers):
    return df[df.columns.intersection(headers)]

def get_test_nums(df) -> Set[int]:
    return {x for x in df.test_no}

pt_test_nums = get_test_nums(pt_results)
reg_test_nums = get_test_nums(reg_results)

def get_rund_ids(df) -> Set[int]:
    return {x for x in df.rund_id}

pt_rund_ids = get_rund_ids(pt_results)
reg_rund_ids = get_rund_ids(reg_results)

PT_NRUNS = len(pt_rund_ids)
REG_NRUNS = len(reg_rund_ids)

print(f"Pre-Trained Runs: {PT_NRUNS}\nRegular Runs: {REG_NRUNS}")

Pre-Trained Runs: 10
Regular Runs: 10

D = {
    "hidden_layers": 2,
    "gen_max": 30,
    "n_runs": 10
}

def validate(D: Dict, df: pd.DataFrame):

    def check_hidden_layers(df: pd.DataFrame, value: str) -> bool:
        return all(df['hidden_layers'] == value)

    print("Correct number of Hidden Layers: ", check_hidden_layers(df, D["hidden_layers"]))


    def check_generations_range(df: pd.DataFrame, minimum: int, maximum: int) -> bool:
        return all(df['generation'] >= minimum) and all(df['generation'] <= maximum)

    print("Correct number of Generations: ", check_generations_range(df, 0, D["gen_max"]))

    def get_ind_rund_ids(df):
        return len({x for x in df.rund_id})

    print("Minimum number of Runs reached: ", get_ind_rund_ids(df) >= D["n_runs"])


print("Regular Results:\n...")
validate(D, reg_results) 
print("\nPre-trained Results:\n...")
validate(D, pt_results)
print()

Regular Results:
...
Correct number of Hidden Layers:  True
Correct number of Generations:  True
Minimum number of Runs reached:  True

Pre-trained Results:
...
Correct number of Hidden Layers:  True
Correct number of Generations:  True
Minimum number of Runs reached:  True

def get_vals(df, vals, gens):

    ret = []

    def filter_df_by_col_val(df, col, val):
        return df[df[col] == val]

    def get_rund_ids(df) -> Set[int]:
        return {x for x in df.rund_id}
    
    run_ids = get_rund_ids(df)

    for i, id in enumerate(run_ids):

        _df = filter_df_by_col_val(df, "rund_id", id)

        ret.append([])

        for gen in range(0,gens+1):

            __df = filter_df_by_col_val(_df, "generation", gen)

            ret[i].append(
                __df[vals].values[0]
            )
    return ret


reg_fits = get_vals(reg_results, "best_fitness", 30)
reg_fits_test = get_vals(reg_results, "best_fitness_test", 30)

pt_fits = get_vals(pt_results, "best_fitness", 30)
pt_fits_test = get_vals(pt_results, "best_fitness_test", 30)


def get_means(arr):

    ret = []

    for gen in range(0, 31):

        gen_fits=[]

        for run in range(0, len(arr)):

            gen_fits.append(arr[run][gen])
    
        ret.append(mean(gen_fits))

    return ret
        


reg_fits_mean = get_means(reg_fits)
reg_fits_test_mean = get_means(reg_fits_test)

pt_fits_mean = get_means(pt_fits)
pt_fits_test_mean = get_means(pt_fits_test)


def get_medians(arr):

    ret = []

    for gen in range(0, 31):

        gen_fits=[]

        for run in range(0, len(arr)):

            gen_fits.append(arr[run][gen])
    
        ret.append(median(gen_fits))

    return ret


reg_fits_med = get_medians(reg_fits)
reg_fits_test_med = get_medians(reg_fits_test)

pt_fits_med = get_medians(pt_fits)
pt_fits_test_med = get_medians(pt_fits_test)

DATAPATH = "/Users/rmn/masterThesis/master_thesis/data/energyCooling_2hl_FullRun_30gens"

def writeMWU(dir_name: str, file_name:str, sample1: Iterable, sample2: Iterable):
    if not os.path.exists(dir_name):
        os.makedirs(dir_name)

    statistic, pval = mannwhitneyu(x=sample1, y=sample2)

    S = {
        "statistic" : statistic,
        "p-value"   : pval
    }
    print(S)

    json.dump(
        S,
        open(os.path.join(dir_name, f"{file_name}.json"), "w", encoding="utf-8"),
    )

def writeData(dir_name: str, file_name:str, D: Dict):
    if not os.path.exists(dir_name):
        os.makedirs(dir_name)


    print(D)

    json.dump(
        D,
        open(os.path.join(dir_name, f"{file_name}.json"), "w", encoding="utf-8"),
    )

from json import load

MPL_CONFIG = load(
    open("/Users/rmn/masterThesis/eda-gp-2020/experiments/matplotlib_config.json", "r", encoding="utf-8")
)


mpl_style.use(MPL_CONFIG["mpl_style"])

# font sizes
SMALL=MPL_CONFIG["fonts"]["small"]
MID=MPL_CONFIG["fonts"]["mid"]
BIG=MPL_CONFIG["fonts"]["big"]

# color codes
C_REG=MPL_CONFIG["colors"]["dae-gp"]
C_PT=MPL_CONFIG["colors"]["pt_dae-gp"]

# marker codes
M_TRAIN=MPL_CONFIG["marker"]["train"]
M_TEST=MPL_CONFIG["marker"]["test"]

TRAIN_LINESTYLE=MPL_CONFIG["train_line_style"]

DPI=MPL_CONFIG["dpi"]


IMG_PATH=f"{MPL_CONFIG['image_base_path']}/energyCooling_2hl_maxIndSize_fullRun_30gens"

def create_dir(dir_name):
    if not os.path.exists(dir_name):
        os.makedirs(dir_name)

create_dir(IMG_PATH)
BASE_TITLE="Energy Cooling 2 Hidden Layer"

fig, (axl, axr) = plt.subplots(ncols=2, layout="constrained", sharex=True, sharey=True, dpi=DPI)
fig.set_size_inches(14,12)
gens = [x for x in range(0, 31)]



fig.suptitle(f"{BASE_TITLE} - Best Fitness by generation", fontsize=BIG)
fig.supxlabel("Generations", fontsize=MID)
fig.supylabel("RMSE", fontsize=MID)

axl.set_title(f"Mean")
axl.plot(gens, reg_fits_mean, color=C_REG, marker=M_TRAIN, linestyle=TRAIN_LINESTYLE, label="DAE-GP(Train)")
axl.plot(gens, reg_fits_test_mean, color=C_REG, marker=M_TEST, label="DAE-GP(Test)")
axl.plot(gens, pt_fits_mean, color=C_PT, marker=M_TRAIN, linestyle=TRAIN_LINESTYLE, label="Pre-Trained(Train)")
axl.plot(gens, pt_fits_test_mean, color=C_PT, marker=M_TEST, label="Pre-Trained(Test)")
axl.grid()

axr.set_title("Median")
axr.plot(gens, reg_fits_med, color=C_REG, marker=M_TRAIN, linestyle=TRAIN_LINESTYLE, label="DAE-GP(Train)")
axr.plot(gens, reg_fits_test_med, color=C_REG, marker=M_TEST, label="DAE-GP(Test)")
axr.plot(gens, pt_fits_med, color=C_PT, marker=M_TRAIN, linestyle=TRAIN_LINESTYLE, label="Pre-Trained(Train)")
axr.plot(gens, pt_fits_test_med, color=C_PT, marker=M_TEST, label="Pre-Trained(Test)")
axr.grid()

axr.legend()


fig.savefig(f"{IMG_PATH}/mean_median_fitness_byGens.png")
writeMWU(DATAPATH, "MWU-BestFitnessByGen", pt_fits[0], reg_fits[0])

{'statistic': 933.0, 'p-value': 9.445869297741274e-13}

fig, ax = plt.subplots(ncols=1, layout="constrained", sharex=True, sharey=True, dpi=DPI)
fig.set_size_inches(14,12)
gens = [x for x in range(0, 31)]



fig.suptitle(f"{BASE_TITLE} - Median Best Fitness by generation", fontsize=BIG)
fig.supxlabel("Generations", fontsize=MID)
fig.supylabel("RMSE", fontsize=MID)


ax.set_title("Median")
ax.plot(gens, reg_fits_med, color=C_REG, marker=M_TRAIN, linestyle=TRAIN_LINESTYLE, label="DAE-GP(Train)")
ax.plot(gens, reg_fits_test_med, color=C_REG, marker=M_TEST, label="DAE-GP(Test)")
ax.plot(gens, pt_fits_med, color=C_PT, marker=M_TRAIN, linestyle=TRAIN_LINESTYLE, label="Pre-Trained(Train)")
ax.plot(gens, pt_fits_test_med, color=C_PT, marker=M_TEST, label="Pre-Trained(Test)")
ax.grid()

ax.legend()


fig.savefig(f"{IMG_PATH}/median_fitness_byGens.png")

def last_fits(arr):
    ret = []
    for run in arr:
        ret.append(run[-1])
    return ret

fig, ax = plt.subplots(layout="constrained", sharex=True, sharey=True, dpi=DPI)
fig.set_size_inches(14,12)

LABELS = ["DAE-GP(Train)", "DAE-GP(Test)", "Pre-Trained(Train)", "Pre-Trained(Test)"]

X = [
    last_fits(reg_fits),
    last_fits(reg_fits_test),
    last_fits(pt_fits),
    last_fits(pt_fits_test)
]

std_dev = np.std(X, 1)
means = np.mean(X,1)

fig.suptitle(f"{BASE_TITLE} - Best Fitness after 30 gens", fontsize=BIG)
fig.supylabel("RMSE", fontsize=MID)

bp_dict = ax.boxplot(
    x=X,
    labels=LABELS,
    #patch_artist=True,  # fill with color
    #notch=True,  # notch shape
)

for i, line in enumerate(bp_dict['medians']):
    x, y = line.get_xydata()[1]
    text = ' mean={:.2f}\n std_dev={:.2f}'.format(means[i], std_dev[i])
    ax.annotate(text, xy=(x, y))


ax.grid()
fig.savefig(f"{IMG_PATH}/final_fit_boxplot.png")

D = {
    "problem": "Energy(Cooling)",
    "hiddenLayer": 2,
    "DAE-GP (train)" : last_fits(reg_fits),
    "DAE-GP (test)" : last_fits(reg_fits_test),
    "Pre-Trained (train)" : last_fits(pt_fits),
    "Pre-Trained (test)" : last_fits(pt_fits_test)
}

writeData(DATAPATH, "best_final_fitness", D)

{'problem': 'Energy(Cooling)', 'hiddenLayer': 2, 'DAE-GP (train)': [4.337181364377099, 4.480445098564427, 4.550882869931211, 3.954563917962384, 4.533227413361258, 4.480445098564427, 4.337181364377099, 4.337104482457938, 5.895299151301257, 4.3649513766860375], 'DAE-GP (test)': [4.747497751008771, 4.612536048865454, 4.453997502244472, 4.518117377339445, 4.341987022330137, 4.612536048865454, 4.747497751008771, 4.423649587055852, 5.594871814885485, 4.647598898445662], 'Pre-Trained (train)': [4.67754072559324, 5.109217776025856, 4.984480288856603, 5.612030279156258, 4.453064838774901, 4.480445098564427, 4.67754072559324, 4.533227413361258, 4.627208931958876, 4.627477517728566], 'Pre-Trained (test)': [5.12388054338864, 5.076970731969272, 5.372248811407255, 5.878967498988803, 4.63897516968996, 4.612536048865454, 5.12388054338864, 4.341987022330137, 4.561784989635818, 4.465011635623137]}

# number of evals per generation


reg_nevals = get_vals(reg_results, "fitness_nevals", 30)
pt_nevals = get_vals(pt_results, "fitness_nevals", 30)


reg_nevals_mean = get_means(reg_nevals)
pt_nevals_mean = get_means(pt_nevals)



fig, ax = plt.subplots(layout="constrained", sharex=True, dpi=DPI)
fig.set_size_inches(14,12)
gens = [x for x in range(0, 31)]


fig.suptitle(f"{BASE_TITLE} - Mean Number of Fitness Evaluations", fontsize=BIG)

fig.supxlabel("Generations", fontsize=MID)

ax.set_ylabel("Number of Fitness Evaluations")
ax.plot(gens, reg_nevals_mean, color=C_REG, label="DAE-GP")
ax.plot(gens, pt_nevals_mean, color=C_PT, label="Pre-Trained")

ax.grid()

ax.legend()


fig.savefig(f"{IMG_PATH}/mean_nevals_byGens.png")

# plot total runtime

#  unique rate and lev diversity per generation

reg_levdiv = get_vals(reg_results, "norm_lev_div", 30)
pt_levdiv = get_vals(pt_results, "norm_lev_div", 30)

reg_levdiv_mean = get_means(reg_levdiv)
pt_levdiv_mean = get_means(pt_levdiv)

reg_ur = get_vals(reg_results, "unique_rate", 30)
pt_ur = get_vals(pt_results, "unique_rate", 30)

reg_ur_mean = get_means(reg_ur)
pt_ur_mean = get_means(pt_ur)



fig, (axl, axr) = plt.subplots(ncols=2, layout="constrained", sharex=True, dpi=DPI)
fig.set_size_inches(14,12)
gens = [x for x in range(0, 31)]

axl.set_ylim(bottom=0, top=1)
axl.set_xlim(left=0, right=30)


fig.suptitle(f"{BASE_TITLE} - Mean Population Diversity by generation", fontsize=BIG)

fig.supxlabel("Generations", fontsize=MID)

axr.set_ylabel("Normalized Levenshtein Edit Distance")
axr.plot(gens, reg_levdiv_mean, color=C_REG, label="DAE-GP")
axr.plot(gens, pt_levdiv_mean, color=C_PT, label="Pre-Trained")


axl.set_ylabel("Unique Rate")
axl.plot(gens, reg_ur_mean, color=C_REG, label="DAE-GP")
axl.plot(gens, pt_ur_mean, color=C_PT, label="Pre-Trained")

axl.grid()
axr.grid()

axr.legend()


fig.savefig(f"{IMG_PATH}/mean_popDiversity_byGens.png")

# plot sample time

# plot avg size


reg_avgsize = get_vals(reg_results, "avg_size", 30)
pt_avgsize = get_vals(pt_results, "avg_size", 30)

reg_bestsize = get_vals(reg_results, "size_best_fitness", 30)
pt_bestsize = get_vals(pt_results, "size_best_fitness", 30)

reg_avgsize_mean = get_means(reg_avgsize)
pt_avgsize_mean = get_means(pt_avgsize)

reg_bestsize_mean = get_means(reg_bestsize)
pt_bestsize_mean = get_means(pt_bestsize)

# reg_avgsize_mean = get_means(reg_avgsize)
# pt_avgsize_mean = get_means(pt_avgsize)

# reg_bestsize_mean = get_means(reg_bestsize)
# pt_bestsize_mean = get_means(pt_bestsize)




fig, (axl) = plt.subplots(layout="constrained", sharex=True, sharey=True, dpi=DPI)
fig.set_size_inches(10,8)
gens = [x for x in range(0, 31)]

# ax.set_ylim(bottom=0)
# ax.set_xlim(left=0)


fig.suptitle(f"{BASE_TITLE} - Solution Size by generation", fontsize=BIG)

fig.supxlabel("Generations", fontsize=MID)

axl.set_ylabel("Mean Tree Size")
axl.plot(gens, reg_bestsize_mean, color=C_REG, label="DAE-GP (Best Solution)")
# axl.plot(gens, reg_avgsize_mean, color=C_REG,linestyle=TRAIN_LINESTYLE, label="DAE-GP (Population average)")

axl.plot(gens, pt_bestsize_mean, color=C_PT, label="Pre-Trained (Best Solution)")
# axl.plot(gens, pt_avgsize_mean, color=C_PT, linestyle=TRAIN_LINESTYLE, label="Pre-Trained (Population average)")



axl.grid()

axl.legend()


fig.savefig(f"{IMG_PATH}/mean_Size_byGens.png")

D = {
    "problem": "Energy(Cooling)",
    "hiddenLayer": 2,
    "DAE-GP" : last_fits(reg_bestsize),
    "Pre-Trained" : last_fits(pt_bestsize)
}

writeData(DATAPATH, "size_best_solution", D)

{'problem': 'Energy(Cooling)', 'hiddenLayer': 2, 'DAE-GP': [4.337181364377099, 4.480445098564427, 4.550882869931211, 3.954563917962384, 4.533227413361258, 4.480445098564427, 4.337181364377099, 4.337104482457938, 5.895299151301257, 4.3649513766860375], 'Pre-Trained': [4.67754072559324, 5.109217776025856, 4.984480288856603, 5.612030279156258, 4.453064838774901, 4.480445098564427, 4.67754072559324, 4.533227413361258, 4.627208931958876, 4.627477517728566]}

import datetime

def print_current_date_and_time():
  now = datetime.datetime.now()
  print(f'Notebook last executed at: {now.strftime("%Y-%m-%d %H:%M:%S")}')

print_current_date_and_time()

Notebook last executed at: 2023-01-12 11:41:11